Ton618 Only Stickers Flux Lo R A

Introduction

The TON618-Only-Stickers-Flux-LoRA model, developed by PrithivMLmods, is designed for generating sticker-like images using text prompts. It employs a LoRA (Low-Rank Adaptation) technique for diffusion models, enabling creative text-to-image transformations.

Architecture

The model is built on the FLUX.1-dev base model from Black Forest Labs and utilizes the Diffusion LoRA template. The architecture includes parameters like a constant LR Scheduler, AdamW optimizer, and specific noise settings. It operates at a network dimension of 64 with an alpha of 32, trained over 15 epochs using 15 high-resolution images.

Training

Training involved using a small dataset with labeling in natural English language and Florence2-en. Key parameters included a noise offset of 0.03, multires noise discount of 0.1, and iterations set to 10. The model is still being trained and may display artifacts or less-than-optimal performance.

Guide: Running Locally

1. Set up Environment: Make sure you have Python and PyTorch installed. Use a GPU-enabled device for efficient processing.
2. Import Libraries:

   import torch
   from pipelines import DiffusionPipeline
   

3. Load Base Model:

   base_model = "black-forest-labs/FLUX.1-dev"
   pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)
   

4. Load LoRA Weights:

   lora_repo = "prithivMLmods/Ton618-Only-Stickers-Flux-LoRA"
   pipe.load_lora_weights(lora_repo)
   

5. Set Device:

   device = torch.device("cuda")
   pipe.to(device)
   

6. Generate Images: Use the trigger word "Only Sticker" in your text prompt.

For optimal performance, consider using cloud GPUs such as AWS EC2, Google Cloud Platform, or Azure.

License

The model is licensed under the CreativeML OpenRAIL-M license, allowing both academic and commercial use with proper attribution.